A Deep Learning Model for Spectrum Sensing in Cognitive Radio Networks.
Masood R. Rashid, Mohammed Ahmed Shakir · Zanin Journal of Science and Engineering · 2026
Cognitive radio has emerged as a promising solution for enhancing wireless spectrum utilization, with spectrum sensing serving as one of its core functions. However, traditional sensing methods such as energy detection and feature-based approaches often face challenges in low signal-to-noise ratio environments and under uncertain conditions. Recent advancements in artificial intelligence and deep learning have introduced data-driven methods that offer improved sensing performance. This study presents a hybrid deep learning model that combines convolutional neural networks, long short-term memory networks, and transformer networks to achieve higher spectrum sensing accuracy. In this framework, convolutional neural network layers extract local spatial features, long short-term memory layers model temporal dependencies, and the Transformer block captures global contextual relationships within the signal sequence. Using the RadioML2016.10B dataset, experimental results show that the proposed model achieves a 99% detection probability at 12dB signal-to-noise ratio, outperforming both traditional methods and deep learning models such as DetectNet, convolutional neural networks long short-term memory and recurrent neural network-gated recurrent unit that achieves detection probability of 84%, 78% and 53% respectively for the same signal-to-noise ratio. These findings demonstrate that the hybrid architecture improves the reliability and robustness of spectrum sensing, especially in low signal-to-noise ratio scenarios.